[go: up one dir, main page]

CN112800141A - An on-demand service aggregation based on RGPS metamodel and its recommendation method - Google Patents

An on-demand service aggregation based on RGPS metamodel and its recommendation method Download PDF

Info

Publication number
CN112800141A
CN112800141A CN202011460443.0A CN202011460443A CN112800141A CN 112800141 A CN112800141 A CN 112800141A CN 202011460443 A CN202011460443 A CN 202011460443A CN 112800141 A CN112800141 A CN 112800141A
Authority
CN
China
Prior art keywords
service
input
output
user
role
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202011460443.0A
Other languages
Chinese (zh)
Inventor
赵一
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Guangdong Ocean University
Original Assignee
Guangdong Ocean University
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Guangdong Ocean University filed Critical Guangdong Ocean University
Priority to CN202011460443.0A priority Critical patent/CN112800141A/en
Publication of CN112800141A publication Critical patent/CN112800141A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
    • G06F16/28—Databases characterised by their database models, e.g. relational or object models
    • G06F16/284—Relational databases
    • G06F16/288—Entity relationship models
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
    • G06F16/28—Databases characterised by their database models, e.g. relational or object models
    • G06F16/284—Relational databases
    • G06F16/285—Clustering or classification
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/90—Details of database functions independent of the retrieved data types
    • G06F16/95—Retrieval from the web
    • G06F16/953—Querying, e.g. by the use of web search engines
    • G06F16/9535—Search customisation based on user profiles and personalisation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/044—Recurrent networks, e.g. Hopfield networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/084—Backpropagation, e.g. using gradient descent

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Databases & Information Systems (AREA)
  • Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • General Health & Medical Sciences (AREA)
  • Computational Linguistics (AREA)
  • Evolutionary Computation (AREA)
  • Biophysics (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Biomedical Technology (AREA)
  • Artificial Intelligence (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

为了解决“互联网+”的应用服务聚合与推荐机制中服务数量剧增、服务组织的无序化和用户需求呈现的多样化问题，通过将角色(Role)‑目标(Goal)‑流程(Process)‑服务(Service)需求元模型的语义关联，满足用户的个性化需求的服务。本发明公开了一种基于RGPS元模型的按需服务聚合及其推荐方法，提出基于RGPS关联网络中面向角色和目标反推的、采用LSTM神经网络对潜在服务的个性化推荐方法，通过“旅行”领域的服务推荐仿真实验，根据QoS阈值作为服务筛选器，来提高方法的精准性及有效性，具有实际的应用价值。

Figure 202011460443

In order to solve the problems of the sharp increase in the number of services, the disorder of service organization and the diversification of user needs in the application service aggregation and recommendation mechanism of "Internet +", the role (Role)-goal (Goal)-process (Process) ‑ Semantic association of service requirements meta-model, services that meet the individual requirements of users. The invention discloses an on-demand service aggregation and a recommendation method based on an RGPS meta-model, and proposes a personalized recommendation method for potential services based on the role and target inversion in the RGPS association network and using the LSTM neural network. "Service recommendation simulation experiments in the field of QoS thresholds are used as service filters to improve the accuracy and effectiveness of the method, which has practical application value.

Figure 202011460443

Description

On-demand service aggregation and recommendation method based on RGPS meta-model
Technical Field
The invention belongs to the technical field of computers, and particularly relates to an on-demand service aggregation and recommendation method of an RGPS meta-model.
Background
The rise of the concept of 'internet +' and customized services enables web services forming the internet to have the characteristics of distribution, modularization, self description and the like. Meanwhile, the user requirements on the internet tend to be personalized and complicated, so the problem of accurate matching of the service and the user requirements becomes more important. However, in the actual recommendation of the Web service, the problems that the user demand is variable, the business process is complex, the service resources cannot adapt to the change of the user demand and the like occur, so that the service recommendation becomes difficult. The existing software engineering technology does not fully research the problem, so that a governance and management method which can be suitable for user demand personalization becomes an important research hotspot and further development direction in the field of service aggregation and service recommendation.
Disclosure of Invention
Aiming at the defects of the prior art, the invention provides an on-demand service aggregation based on an RGPS meta-model and a recommendation method thereof.
The invention adopts the following technical scheme:
1. an on-demand service aggregation and recommendation method based on an RGPS meta-model is characterized by comprising the following steps:
step 1, according to the target set related in the specific field problem, firstly adding the Association with the specific targetSDIThe role of the relationship and the association between the two, then the relationship between the role and the target is added, and the ontology (O), the role and the target model (R) are respectively extracted&G) Common core metadata and management information of the process model (P) and the service model (S) establish corresponding SDIs for different types of models so as to promote cross-domain or cross-system query and multi-granularity multiplexing of heterogeneous information models;
step 2, constructing corresponding SDI according to different types of models obtained in the step 1, wherein the realization method is as follows,
a user demand driven RGPS associated network generation step is given, wherein
Figure RE-GDA0003016899700000011
Evaluation of the service (QoS value) if
Figure RE-GDA0003016899700000012
If the satisfaction degree of the user to the service is above the minimum QoS threshold value, the service of role reverse thrust meets the requirement of the user, and the weighted edge between the service and the role node is connected; in the same way, if
Figure RE-GDA0003016899700000013
Representing the user pairIf the satisfaction degree of the service is above the minimum QoS threshold, the service of the target reverse thrust meets the requirement of the user, and the weighted edge between the service and the target node is connected;
and 3, analyzing the user requirements and recommending related potential services by utilizing R and G reverse-thrust service ways and combining an LSTM neural network model, so that new instant services can be recommended for the user.
2. The on-demand service aggregation and recommendation method based on RGPS meta-model as claimed in claim 1, wherein: in step 2, a user demand driven RGPS associated network generation step is given, wherein
Figure RE-GDA0003016899700000021
Evaluation of the service (QoS value) if
Figure RE-GDA0003016899700000022
If the satisfaction degree of the user to the service is above the minimum QoS threshold value, the service of role reverse thrust meets the requirement of the user, and the weighted edge between the service and the role node is connected; in the same way, if
Figure RE-GDA0003016899700000023
If the satisfaction degree of the user to the service is above the minimum QoS threshold value, the service of the target reverse thrust meets the requirement of the user, and the weighted edge between the service and the target node is connected; the RGPS associated network is continuously modified through the feedback correction mechanism, so that the related services which can well meet the requirements of users are found out, the searched service sets can be sequenced, and scientific basis is provided for the customized recommendation of the services.
3. The on-demand service aggregation and recommendation method based on RGPS meta-model as claimed in claim 1, wherein: the key point of expressing aggregated services by using a directed graph model is how to obtain accurate RGPS elements from the demands, namely, the demands are analyzed and modeled by using an LSTM neural network.
The LSTM neural network processes the core idea of service recommendation: the core of the LSTM is "cell state", which functions as a memory space in the whole model, and the memory space changes with time, and the memory space cannot control which users' demand information is memorized, so that it is a control gate that really plays a control role.
(1) The input node receives the output of the hidden node of the previous time point and the current input as the input, and then passes through an activation function of tanh;
(2) an input gate: the gate input is the output of the hidden node at the last time point and the current input, and the activation function is sigmoid (the reason is that the output of the sigmoid is between 0 and 1, and the multiplication of the output of the input gate and the output of the input node can play a role in controlling the information quantity);
(3) internal state node: the input is the current input filtered by the input gate and the internal state node output of the previous time point;
(4) forgetting to remember the door: the function of controlling the internal state information is realized, and the Gate is opened (set to 1) or closed (set to 0) through the training parameters to protect the Cell;
(5) an output gate: the function of controlling output information is realized, the input of the gate is the output of the hidden node at the last time point and the current input, and the activation function is sigmoid;
for example, in the field of 'travel', a user inputs a demand 'i want to reserve a train ticket on the day, reserve a lodging and provide urban traffic information'; the example sentences are subjected to machine recognition and user requirements are marked off, each vertical rectangular frame of the LSTM model in the attached figure 2 represents a hidden layer of each iteration, each hidden layer is provided with a plurality of neurons, each neuron performs linear matrix operation on an input vector, and an output result (for example, a tanh () function) is generated through nonlinear operation. In each iteration, the output of the previous iteration varies with the word vector of the next word in the document, XtIs the input to the hidden layer and the hidden layer will produce the predicted output value and the output feature vector H that is provided to the hidden layer of the next layert。
The expression formula of each layer model of the LSTM neural network is as follows:
an input gate: the function is to control the output of the cell state:
Figure RE-GDA0003016899700000031
memory cell: for the time t-1 → t, the memory cell information change process is represented as t-1 time state and is obtained by filtering with forgetting gate
Figure RE-GDA0003016899700000032
The stored information and the input door information acquired at the time t are compared
Figure RE-GDA0003016899700000033
Addition gives the transition in cell state:
Figure RE-GDA0003016899700000034
output of cell status:
Figure RE-GDA0003016899700000035
then the inverse of the residual is conducted to the output gate to modify the function:
Figure RE-GDA0003016899700000036
the following error propagates to the cell state:
Figure RE-GDA0003016899700000037
error passes to forget gate:
Figure RE-GDA0003016899700000038
after the residual conduction is finished, the residual is directly derived from the weight
Figure RE-GDA0003016899700000039
The invention has the beneficial effects that:
(1) the user requirements are divided into specific fields, then modeling is carried out through common requirements, analysis is carried out on the aspects of roles, targets and processes related to the user requirements, a corresponding on-demand service aggregation algorithm is designed, and an associated network diagram is drawn.
(2) 2 service searching and recommending methods are designed to meet different requirement expression forms provided by users, and the problem that the users recommend a proper potential service set by using cooperation among services is better solved, so that the requirements are met.
(3) A specific experiment is designed to verify and analyze service searching timeliness, precision, recall rate and F value in four aspects, and relevant definitions and algorithms are verified according to specific fields, so that the reliability of the method is proved by quantitative expression.
Drawings
Figure 1 is a diagram of a service aggregation and recommendation framework based on RGPS on-demand guidance.
FIG. 2 is a diagram of an LSTM neural network analysis model for user demand.
Detailed Description
Aiming at the characteristic of disorder of traditional Service aggregation, the invention provides a R, G, P, S-associated weighted network method for orderly organizing Service aggregation based on the semantic association relationship of a Role (Role) -target (Goal) -Process (Process) -Service (Service) requirement meta model. The following description of the embodiments of the present invention will be made with reference to the accompanying drawings: an on-demand service aggregation and recommendation method based on RGPS meta-model is characterized by comprising the following steps:
step 1, according to the target set related in the specific field problem, firstly adding the Association with the specific targetSDIThe role of the relation and the association between the two, then the relation between the role, the role and the target is addedRespectively extracting an ontology (O), a role and a target model (R)&G) Common core metadata and management information of the process model (P) and the service model (S) establish corresponding SDIs for different types of models so as to promote cross-domain or cross-system query and multi-granularity multiplexing of heterogeneous information models;
step 2, constructing corresponding SDI according to different types of models obtained in the step 1, wherein the realization method is as follows,
a user demand driven RGPS associated network generation step is given, wherein
Figure RE-GDA0003016899700000041
Evaluation of the service (QoS value) if
Figure RE-GDA0003016899700000042
If the satisfaction degree of the user to the service is above the minimum QoS threshold value, the service of role reverse thrust meets the requirement of the user, and the weighted edge between the service and the role node is connected; in the same way, if
Figure RE-GDA0003016899700000043
If the satisfaction degree of the user to the service is above the minimum QoS threshold value, the service of the target reverse thrust meets the requirement of the user, and the weighted edge between the service and the target node is connected;
and 3, analyzing the user requirements and recommending related potential services by utilizing R and G reverse-thrust service ways and combining an LSTM neural network model, so that new instant services can be recommended for the user.
2. The on-demand service aggregation and recommendation method based on RGPS meta-model as claimed in claim 1, wherein: in step 2, a user demand driven RGPS associated network generation step is given, wherein
Figure RE-GDA0003016899700000044
Evaluation of the service (QoS value) if
Figure RE-GDA0003016899700000045
Is shown byThe satisfaction degree of the user to the service is above the minimum QoS threshold value, and if the service of role reverse thrust meets the requirement of the user, the weighted edge between the service and the role node is connected; in the same way, if
Figure RE-GDA0003016899700000046
If the satisfaction degree of the user to the service is above the minimum QoS threshold value, the service of the target reverse thrust meets the requirement of the user, and the weighted edge between the service and the target node is connected; the RGPS associated network is continuously modified through the feedback correction mechanism, so that the related services which can well meet the requirements of users are found out, the searched service sets can be sequenced, and scientific basis is provided for the customized recommendation of the services.
3. The on-demand service aggregation and recommendation method based on RGPS meta-model as claimed in claim 1, wherein: the key point of expressing aggregated services by using a directed graph model is how to obtain accurate RGPS elements from the demands, namely, the demands are analyzed and modeled by using an LSTM neural network.
The LSTM neural network processes the core idea of service recommendation: the core of the LSTM is "cell state", which functions as a memory space in the whole model, and the memory space changes with time, and the memory space cannot control which users' demand information is memorized, so that it is a control gate that really plays a control role.
(1) The input node receives the output of the hidden node of the previous time point and the current input as the input, and then passes through an activation function of tanh;
(2) an input gate: the gate input is the output of the hidden node at the last time point and the current input, and the activation function is sigmoid (the reason is that the output of the sigmoid is between 0 and 1, and the multiplication of the output of the input gate and the output of the input node can play a role in controlling the information quantity);
(3) internal state node: the input is the current input filtered by the input gate and the internal state node output of the previous time point;
(4) forgetting to remember the door: the function of controlling the internal state information is realized, and the Gate is opened (set to 1) or closed (set to 0) through the training parameters to protect the Cell;
(5) an output gate: the function of controlling output information is realized, the input of the gate is the output of the hidden node at the last time point and the current input, and the activation function is sigmoid;
for example, in the field of 'travel', a user inputs a demand 'i want to reserve a train ticket on the day, reserve a lodging and provide urban traffic information'; the example sentences are subjected to machine recognition and user requirements are marked off, each vertical rectangular frame of the LSTM model in the attached figure 2 represents a hidden layer of each iteration, each hidden layer is provided with a plurality of neurons, each neuron performs linear matrix operation on an input vector, and an output result (for example, a tanh () function) is generated through nonlinear operation. In each iteration, the output of the previous iteration varies with the word vector of the next word in the document, XtIs the input to the hidden layer and the hidden layer will produce the predicted output value and the output feature vector H that is provided to the hidden layer of the next layert。
The expression formula of each layer model of the LSTM neural network is as follows:
an input gate: the function is to control the output of the cell state:
Figure RE-GDA0003016899700000051
memory cell: for the time t-1 → t, the memory cell information change process is represented as t-1 time state and is obtained by filtering with forgetting gate
Figure RE-GDA0003016899700000052
The stored information and the input door information acquired at the time t are compared
Figure RE-GDA0003016899700000053
Addition gives the transition in cell state:
Figure RE-GDA0003016899700000061
output of cell status:
Figure RE-GDA0003016899700000062
then the inverse of the residual is conducted to the output gate to modify the function:
Figure RE-GDA0003016899700000063
the following error propagates to the cell state:
Figure RE-GDA0003016899700000064
error passes to forget gate:
Figure RE-GDA0003016899700000065
after the residual conduction is finished, the residual is directly derived from the weight
Figure RE-GDA0003016899700000066
Example 1
The following are specific examples of the application of the present invention:
in the specific field of travel, accurate service recommendation is carried out on user requirements, and the method comprises the following steps: user requirements are identified according to the algorithm and a correlation network is established, semantic decomposition is carried out on the user requirements through a requirement acquisition and analysis tool developed by the subject group, 4 roles corresponding to the requirements are provided, namely 'train passenger', 'accommodation person', 'consultant' and 'tourist', and 5 targets are respectively: "order train tickets", "booking hotels", "urban traffic", "urban weather", "urban scenic spots";
step 1 is executed, and the process of labeling the user requirement comprises the following steps: the label for "train passenger" is labeled "1", the label for "accommodation" is labeled "2", the label for "consultant" is labeled "3", and the label for "lesson" is labeled "4". The process of labeling the target is to label the label targeting "order train tickets" as "1", label targeting "booking hotels" as "2", label targeting "urban traffic" as "3", label targeting "urban weather" as "4", and label targeting "urban attractions" as "5".
Step 2 is executed, if the goal is realized, the process decomposition can be carried out, for example, the process of ordering the train ticket can be decomposed into two processes of inquiring the ticket and purchasing the ticket; "booking a hotel" may be broken down into "querying a hotel", "online payment system"; "urban traffic" can be decomposed into "calling maps", "navigation route generation"; "City weather" can be broken down into "Call weather forecast"; the "city attractions" can be broken down into "generate hot attractions" and "ticket reservations".
Step 3 is executed, the service recommendation process is started by the role layer, the purpose of the train passenger is to order the train ticket, the process flow is the steps of inquiring the ticket price and purchasing the ticket, and the sub-process of purchasing the ticket can be divided into providing the ticket information and paying online. Through roles, targets and processes, the services can be classified accurately, through a sequential logic diagram of an RGPS (geographic grouping service) association network, in order to recommend better services such as 'tourist' roles to users, recommended sub-services can be selected from three service clusters of 'payment system', 'navigation software' and 'travel services', and the method can effectively improve the rationality of service recommendation.
After step 3, the model is evaluated with the test set after the model parameters are substantially fixed.
It is to be understood that the above description is not intended to limit the present invention, and the present invention is not limited to the above examples, and those skilled in the art may make modifications, alterations, additions or substitutions within the spirit and scope of the present invention.

Claims (3)

1.一种基于RGPS元模型的按需服务聚合及其推荐方法，其特征在于，包含以下步骤：1. an on-demand service aggregation based on RGPS metamodel and its recommendation method, is characterized in that, comprises the following steps: 步骤1，根据具体领域问题中涉及的目标集合，首先加入与特定目标AssociationSDI关系的角色以及两者之间的关联，然后添加角色与角色及目标之间的关系，分别抽取本体(O)、角色与目标模型(R&G)、过程模型(P)和服务模型(S)的公共核心元数据和管理信息，为不同类型的模型构建了相应的SDI，以促进异构信息模型的跨域或跨系统查询以及多粒度的复用；Step 1: According to the target set involved in the specific domain problem, first add the role and the relationship between the specific target Association SDI relationship and the relationship between the two, and then add the relationship between the role, the role and the target, and extract the ontology (O), Common core metadata and management information for Role and Goal Models (R&G), Process Models (P) and Service Models (S), corresponding SDIs are constructed for different types of models to facilitate cross-domain or cross-distribution of heterogeneous information models System query and multi-granularity reuse; 步骤2，根据采用步骤1所得不同类型的模型构建了相应的SDI，实现方式如下，In step 2, the corresponding SDI is constructed according to the different types of models obtained in step 1, and the implementation method is as follows: 给出了用户需求驱动的RGPS关联网络生成步骤，其中
Figure RE-FDA0003016899690000011
对服务的评价(QoS值)，若
Figure RE-FDA0003016899690000012
时，表示用户对服务的满意度在最低QoS阈值之上，说明角色反推的服务满足用户的需求，则连接服务与角色节点之间的带权边；同理若
Figure RE-FDA0003016899690000013
时，表示用户对服务的满意度在最低QoS阈值之上，说明目标反推的服务满足用户的需求，则连接服务与目标节点之间的带权边；The steps of generating the RGPS association network driven by user requirements are given, in which
Figure RE-FDA0003016899690000011
The evaluation of the service (QoS value), if
Figure RE-FDA0003016899690000012
When , it means that the user’s satisfaction with the service is above the minimum QoS threshold, indicating that the role-reverse service meets the user’s needs, then connect the weighted edge between the service and the role node; similarly, if
Figure RE-FDA0003016899690000013
When , it means that the user's satisfaction with the service is above the minimum QoS threshold, indicating that the target-reverse service meets the user's needs, then connect the weighted edge between the service and the target node; 步骤3，利用R和G反推服务的途径，结合LSTM神经网络模型，来解析用户需求并推荐相关的潜在服务，因此可以为用户推荐新的即时服务。Step 3, using R and G to infer services, combined with the LSTM neural network model, to analyze user needs and recommend related potential services, so new instant services can be recommended for users. 2.根据权利要求书1所述一种基于RGPS元模型的按需服务聚合及其推荐方法，其特征在于：在步骤2中，给出了用户需求驱动的RGPS关联网络生成步骤，其中
Figure RE-FDA0003016899690000016
对服务的评价(QoS值)，若
Figure RE-FDA0003016899690000014
时，表示用户对服务的满意度在最低QoS阈值之上，说明角色反推的服务满足用户的需求，则连接服务与角色节点之间的带权边；同理若
Figure RE-FDA0003016899690000015
时，表示用户对服务的满意度在最低QoS阈值之上，说明目标反推的服务满足用户的需求，则连接服务与目标节点之间的带权边；通过这种反馈修正机制对RGPS关联网络进行不断地修改，从而找出能够很好符合用户需要的相关服务并能够对搜索出的服务集合进行排序，为服务的定制化推荐提供了科学的依据。2. a kind of on-demand service aggregation based on RGPS metamodel according to claim 1 and its recommendation method, it is characterized in that: in step 2, the RGPS associated network generation step driven by user requirements is provided, wherein
Figure RE-FDA0003016899690000016
The evaluation of the service (QoS value), if
Figure RE-FDA0003016899690000014
When , it means that the user’s satisfaction with the service is above the minimum QoS threshold, indicating that the role-reverse service meets the user’s needs, then connect the weighted edge between the service and the role node; similarly, if
Figure RE-FDA0003016899690000015
When , it means that the user’s satisfaction with the service is above the minimum QoS threshold, indicating that the target-reverse service meets the user’s needs, then connect the weighted edge between the service and the target node; through this feedback correction mechanism, the RGPS associated network is connected. Continuous modification is made to find relevant services that can well meet the needs of users and to sort the searched service set, which provides a scientific basis for customized service recommendation. 3.根据权利要求书1所述一种基于RGPS元模型的按需服务聚合及其推荐方法，其特征在于：采用有向图模型来表达聚合的服务，其关键点是如何从需求中获得准确的RGPS元素，即利用LSTM神经网络对需求进行分析建模。3. a kind of on-demand service aggregation based on RGPS metamodel according to claim 1 and its recommendation method, it is characterized in that: adopt directed graph model to express the service of aggregation, and its key point is how to obtain accurate from demand The RGPS element of the LSTM neural network is used to analyze and model the requirements. LSTM神经网络处理服务推荐的核心思想：LSTM的核心是“细胞状态(cell state)”它的功能就是整个模型中的记忆空间，它随着时间而变化，记忆空间无法控制哪些用户的需求信息是否被记忆，所以真正起控制作用的是控制门。The core idea of LSTM neural network processing service recommendation: The core of LSTM is "cell state". Its function is the memory space in the entire model. It changes with time, and the memory space cannot control which user's demand information is is memorized, so what really controls is the control gate. (1)输入节点:输入节点接受上一个时刻点的隐藏节点的输出以及当前的输入作为输入，然后通过一个tanh的激活函数；(1) Input node: The input node accepts the output of the hidden node at the previous time point and the current input as input, and then passes through a tanh activation function; (2)输入门：起控制输入信息的作用，门的输入为上一个时刻点的隐藏节点的输出以及当前的输入，激活函数为sigmoid(原因为sigmoid的输出为0-1之间，将输入门的输出与输入节点的输出相乘可以起控制信息量的作用)；(2) Input gate: It plays the role of controlling the input information. The input of the gate is the output of the hidden node at the previous time point and the current input, and the activation function is sigmoid (the reason is that the output of sigmoid is between 0 and 1, and the input The output of the gate is multiplied by the output of the input node to control the amount of information); (3)内部状态节点：输入为被输入门过滤后的当前输入以及前一时间点的内部状态节点输出；(3) Internal state node: the input is the current input filtered by the input gate and the output of the internal state node at the previous time point; (4)忘记门：起控制内部状态信息的作用，通过训练参数，将Gate或开(置1)或闭(置0)，保护Cell；(4) Forgetting the gate: It plays the role of controlling the internal state information. Through the training parameters, the Gate is opened (set to 1) or closed (set to 0) to protect the Cell; (5)输出门：起控制输出信息的作用，门的输入为上一个时刻点的隐藏节点的输出以及当前的输入，激活函数为sigmoid；(5) Output gate: plays the role of controlling the output information, the input of the gate is the output of the hidden node at the previous time point and the current input, and the activation function is sigmoid; 例如“旅行”领域中，用户输入需求“本人希望预定当天的火车票，同时需要预定住宿，还希望提供城市交通信息”；我们对上述例句进行机器识别并划分出用户需求，附图2中LSTM模型各个垂直矩形框代表每轮迭代的隐层，每个这样的隐层都拥有若干神经元，每个神经元都对输入向量执行线性矩阵操作，通过非线性操作产生输出结果(例如，tanh()函数)。在每一轮迭代中，前一步迭代的输出随着文档中下一条词汇的词向量而变化，Xt是隐层的输入且隐层将产生预测输出值和提供给下一层隐层的输出特征向量Ht。For example, in the field of "travel", the user enters the requirement "I want to book a train ticket for the day, and also need to book accommodation, and also want to provide urban traffic information"; we machine recognize the above example sentences and divide the user needs, the LSTM in Figure 2 Each vertical rectangle of the model represents the hidden layer of each iteration. Each such hidden layer has several neurons. Each neuron performs linear matrix operations on the input vector and produces output results through nonlinear operations (for example, tanh( )function). In each iteration, the output of the previous iteration changes with the word vector of the next word in the document, X t is the input of the hidden layer and the hidden layer will produce the predicted output value and the output provided to the next hidden layer Eigenvector H t . LSTM神经网络每层模型的表达公式为：The expression formula of each layer model of LSTM neural network is: 输入门：作用是控制cell state的输出：Input gate: The function is to control the output of the cell state:
Figure RE-FDA0003016899690000021
Figure RE-FDA0003016899690000021
 记忆细胞：对于时刻t-1→t,记忆细胞信息变化过程表示为t-1时刻状态用遗忘门过滤得到
Figure RE-FDA0003016899690000022
将保存的信息与t时刻获取的输入门信息
Figure RE-FDA0003016899690000023
相加得到细胞状态的转变：Memory cells: For time t-1→t, the information change process of memory cells is expressed as the state at time t-1 and filtered by forgetting gate.
Figure RE-FDA0003016899690000022
Compare the saved information with the input gate information obtained at time t
Figure RE-FDA0003016899690000023
Add up to get the transition of the cell state:
Figure RE-FDA0003016899690000024
Figure RE-FDA0003016899690000024
 细胞状态的输出：
Figure RE-FDA0003016899690000025
Output of cell state:
Figure RE-FDA0003016899690000025
 然后是残差的反向传导至输出门来修正函数：The residuals are then back-propagated to the output gate to modify the function:
Figure RE-FDA0003016899690000031
Figure RE-FDA0003016899690000031
 接下来误差传播到细胞状态：Next the error propagates to the cell state:
Figure RE-FDA0003016899690000032
Figure RE-FDA0003016899690000032
 误差传到遗忘门:The error is passed to the forget gate:
Figure RE-FDA0003016899690000033
Figure RE-FDA0003016899690000033
 残差传导完成后，直接对残差对权值求导得After the residual transmission is completed, the residual is directly derived from the weight to obtain
Figure RE-FDA0003016899690000034
。
Figure RE-FDA0003016899690000034
.
CN202011460443.0A 2020-12-11 2020-12-11 An on-demand service aggregation based on RGPS metamodel and its recommendation method Pending CN112800141A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202011460443.0A CN112800141A (en) 2020-12-11 2020-12-11 An on-demand service aggregation based on RGPS metamodel and its recommendation method

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202011460443.0A CN112800141A (en) 2020-12-11 2020-12-11 An on-demand service aggregation based on RGPS metamodel and its recommendation method

Publications (1)

Publication Number Publication Date
CN112800141A true CN112800141A (en) 2021-05-14

Family

ID=75806328

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202011460443.0A Pending CN112800141A (en) 2020-12-11 2020-12-11 An on-demand service aggregation based on RGPS metamodel and its recommendation method

Country Status (1)

Country Link
CN (1) CN112800141A (en)

Citations (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20080005055A1 (en) * 2006-06-30 2008-01-03 Microsoft Corporation Methods and architecture for learning and reasoning in support of context-sensitive reminding, informing, and service facilitation
US20110282814A1 (en) * 2010-05-12 2011-11-17 Salesforce.Com, Inc. Methods and systems for implementing a compositional recommender framework
CN102880725A (en) * 2012-10-23 2013-01-16 武汉大学 Recommending method based on demand-based service organization
CN106209959A (en) * 2015-05-26 2016-12-07 徐尚英 Network service intelligence based on user's request finds method
CN106650825A (en) * 2016-12-31 2017-05-10 中国科学技术大学 Automotive exhaust emission data fusion system
US20170220966A1 (en) * 2016-02-03 2017-08-03 Operr Technologies, Inc. Method and System for On-Demand Customized Services
CN108446021A (en) * 2018-02-28 2018-08-24 天津大学 Application process of the P300 brain-computer interfaces in smart home based on compressed sensing
CN109343505A (en) * 2018-09-19 2019-02-15 太原科技大学 Prediction method of gear remaining life based on long short-term memory network

Patent Citations (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20080005055A1 (en) * 2006-06-30 2008-01-03 Microsoft Corporation Methods and architecture for learning and reasoning in support of context-sensitive reminding, informing, and service facilitation
US20110282814A1 (en) * 2010-05-12 2011-11-17 Salesforce.Com, Inc. Methods and systems for implementing a compositional recommender framework
CN102880725A (en) * 2012-10-23 2013-01-16 武汉大学 Recommending method based on demand-based service organization
CN106209959A (en) * 2015-05-26 2016-12-07 徐尚英 Network service intelligence based on user's request finds method
US20170220966A1 (en) * 2016-02-03 2017-08-03 Operr Technologies, Inc. Method and System for On-Demand Customized Services
CN106650825A (en) * 2016-12-31 2017-05-10 中国科学技术大学 Automotive exhaust emission data fusion system
CN108446021A (en) * 2018-02-28 2018-08-24 天津大学 Application process of the P300 brain-computer interfaces in smart home based on compressed sensing
CN109343505A (en) * 2018-09-19 2019-02-15 太原科技大学 Prediction method of gear remaining life based on long short-term memory network

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
YI ZHAO ET AL: "An on-demand service aggregation and service recommendation method based on RGPS", 《INTELLIGENT DATA ANALYSIS》 *
刘建晓 等: "RGPS制导的按需服务组织与推荐方法", 《计算机学报》 *
赵一: "基于领域知识的服务聚类与个性化推荐方法", 《中国优秀博硕士学位论文全文数据库（博士）信息科技辑》 *

Similar Documents

Publication Publication Date Title
Bi et al. Daily tourism volume forecasting for tourist attractions
Awoke et al. Bitcoin price prediction and analysis using deep learning models
CN110633409B (en) Automobile news event extraction method integrating rules and deep learning
Zhene et al. Deep convolutional mesh RNN for urban traffic passenger flows prediction
CN104679743A (en) Method and device for determining preference model of user
Liu et al. Unified route representation learning for multi-modal transportation recommendation with spatiotemporal pre-training
Ullah et al. Deep edu: A deep neural collaborative filtering for educational services recommendation
CN108681739A (en) One kind recommending method based on user feeling and time dynamic tourist famous-city
Zhang et al. A spatiotemporal graph generative adversarial networks for short-term passenger flow prediction in urban rail transit systems
Boppana et al. Web crawling based context aware recommender system using optimized deep recurrent neural network
Mo et al. Cross-city multi-granular adaptive transfer learning for traffic flow prediction
CN112487109A (en) Entity relationship extraction method, terminal and computer readable storage medium
Brahimi et al. Modelling on Car‐Sharing Serial Prediction Based on Machine Learning and Deep Learning
Garrido-Munoz et al. A holistic approach for image-to-graph: application to optical music recognition
Osojnik et al. Incremental predictive clustering trees for online semi-supervised multi-target regression
Xian et al. A multi-modal time series intelligent prediction model
Caschera et al. MONDE: a method for predicting social network dynamics and evolution
Wang et al. DeepFM-based taxi pick-up area recommendation
Hu et al. An attention-mechanism-based traffic flow prediction scheme for smart city
CN112800141A (en) An on-demand service aggregation based on RGPS metamodel and its recommendation method
Wei et al. Traffic flow prediction with multi-feature spatio-temporal coupling based on peak time embedding
Li et al. Umformer: a transformer dedicated to univariate multistep prediction
CN114298023B (en) User decision demand generation method and application based on task subject word driving
Liang et al. Combining individual travel preferences into destination prediction: A multi-module deep learning network
Lu et al. Research on sample selection of urban rail transit passenger flow forecasting based on SCBP algorithm

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
WD01 Invention patent application deemed withdrawn after publication
WD01 Invention patent application deemed withdrawn after publication

Application publication date: 20210514